Scenario
The Taguchi method found the optimal parameters: Material temperature 220°C, Injection speed Medium, Packing pressure 60%. Predicted strength = 48.5 N, current process is 42.3 N, a very good improvement.
Your supervisor reviewed it and said: "Okay, let's proceed with mass production using this setting."
You should first say: "Wait a moment, we need to run a confirmation experiment first."
Why a Confirmation Run is Needed
DOE conclusions are predicted values extrapolated from a limited number of experimental points, not a guarantee for mass production. The question a confirmation run answers is:
"Do the actual results obtained under optimal parameters align with the DOE predicted values?"
If inconsistent, it indicates:
- There are interactions not captured by the DOE.
- Important noise factors (environment, operator, batch) are having an influence.
- The experimental range of the DOE is not accurate enough.
How to Conduct a Confirmation Run
Step 1: Run 3-5 repeated experiments under the optimal parameters.
Do not run it only once. The process itself has variability, and a single run might coincidentally be good or bad.
Step 2: Calculate the mean and confidence interval for the confirmation run.
The calculation of the Confidence Interval (CI) requires the error estimate from the DOE:
CI = ŷ ± t × √(MSₑ × (1/nₑff + 1/nconfirm))
Where nₑff is the effective number of replicates, calculated from the DOE design.
Step 3: Determine if the predicted value falls within the confidence interval of the confirmation run.
- ✅ Predicted value falls within CI: DOE conclusion is reliable, can proceed to mass production.
- ❌ Predicted value not within CI: Need to re-examine the DOE to find missing factors.
Significance of Passing and Failing
Passing (Predicted value within CI):
- DOE model is reliable.
- New SOP can be formulated based on optimal parameters.
- Enter the Control phase of DMAIC.
Failing (Predicted value not within CI):
Don't be discouraged; it means you've discovered new information:
- Possible interactions: Consider conducting a full factorial experiment.
- Possible noise factors: Redesign using the S/N ratio method.
- Possible issues with experiment execution: Confirm consistent environmental conditions for each experiment.
Common Mistakes
| Mistake | Consequence |
|---|---|
| Skipping confirmation runs and going directly to mass production | Discovering DOE conclusions are unreliable only after mass production. |
| Running only 1 confirmation run | Unable to distinguish between improvement effects and random variation. |
| Confirmation experiment conditions differ from DOE | Conclusions are not comparable. |
| Giving up on DOE after a failed confirmation | Missing an opportunity to discover new problems. |
Position of Confirmation Run in DMAIC
The confirmation run is the final step in the Improve phase, a necessary gate before entering Control. It serves as a bridge from "laboratory discovery" to "factory implementable."
Golden Quote
"The optimal solution found by DOE is a destination on a map; the confirmation run is the step that verifies the road truly leads there. To proceed to mass production without confirmation is to mistake the map for reality."